Datasets:

HuggingFaceH4

/

no_robots

like 447

Follow

Hugging Face H4 643

I totally agree! It would be great to make it commercial!

Agreed!

yes - please cc-by it or at least bloom-ify it.

Yes, please license it under a permissive license, like MIT, like Zephyr!

Hello everyone, we'd also have liked to release No Robots with a permissive license, but the NC restriction was a requirement from the vendor in order to make the dataset publicly available. Nevertheless, we hope the dataset will still serve as a useful artifact to the community and enable more research on understanding the role of synthetic vs human produced data for SFT models!

Hi, thanks for this info. Could you share the name of the vendor you got this data from? Also, can Hugging Face use this for commercial purposes?

The dataset has numerous errors in it. I came here to report them but seeing as the dataset is non-commercial I guess I'll just throw it all away instead, as I can't use it.

How does this compare to OASST? OASST is under Apache 2.0 license.

Looking forward to commercial use soon!

I don't think anyone is giving anyone a hard time on purpose, we are merely expressing our opinions that the dataset would be far more useful if it were commercial use.

@mrfakename Does OASST sound like GPTs? And what about this dataset?

Hi, thanks for this info. Could you share the name of the vendor you got this data from? Also, can Hugging Face use this for commercial purposes?

Hi @mrfakename , the vendor requested anonymity so I cannot share this information unfortunately.

Thanks. Would you mind sharing more details? For example, although the vendor requested anonymity, is it possible to purchase this dataset elsewhere? Or is this dataset exclusive? And did Hugging Face hire people to create this dataset?

Thanks. Would you mind sharing more details? For example, although the vendor requested anonymity, is it possible to purchase this dataset elsewhere? Or is this dataset exclusive? And did Hugging Face hire people to create this dataset?

You can certainly purchase similar SFT datasets from the vendors listed in the OpenAI / Anthropic / Llama papers - this was how we initially narrowed down the search space :) We didn't hire the annotators directly for the dataset; instead we explained to the vendor what distribution of tasks we wanted and that we didn't want any LLM-generated outputs. Hope that helps!

Hi, since this is NC-licensed, does this also apply to trained models? For example, if I trained a model on this dataset, would it have to be NC as well?

@mrfakename IANAL but the prevailing view until now is that model licensing is independent from the datasets the model has been trained on

Thanks. Would you mind sharing more details? For example, although the vendor requested anonymity, is it possible to purchase this dataset elsewhere? Or is this dataset exclusive? And did Hugging Face hire people to create this dataset?

You can certainly purchase similar SFT datasets from the vendors listed in the OpenAI / Anthropic / Llama papers - this was how we initially narrowed down the search space :) We didn't hire the annotators directly for the dataset; instead we explained to the vendor what distribution of tasks we wanted and that we didn't want any LLM-generated outputs. Hope that helps!

Dear Lewis, thank you for the work on the No Robots dataset. Will a technical report with more information on the prompt creation process follow? For a low resource language I would like to replicate the annotation procedure and am for this reason curious how the prompts were selected and created, which quality control measures were in place and what tooling has been used to perform the annotations.